Review Velocity and AI Citation Threshold: Research Methodology for Indian Local Businesses

Research Questions

This study investigates three primary research questions:

RQ1: What is the minimum Google review count (the "AI citation threshold") at which businesses in each category × city combination begin appearing in Google AI Overviews for recommendation queries?

RQ2: Does monthly review velocity affect AI Overview citation rates independently of total review count (i.e., do two businesses with the same total review count but different monthly velocities receive different citation rates)?

RQ3: How long after crossing the review count threshold does a business begin appearing consistently in AI Overviews?

Status: this is a research protocol. No threshold number, velocity effect, or lag figure below is a finding — every number in this document is a description of what will be measured, not what was measured. There is no published, universal review-count threshold that unlocks AI citation anywhere in the industry, and this study is not designed to produce one either; it's designed to produce category- and city-specific observed floors, which are a different and more honest kind of number.

Why this question matters to a real decision

Every business owner building review volume asks some version of the same question: how many reviews is "enough," and does it matter how fast they come in versus how many total exist. That question currently gets answered with folklore — a competitor mentioned a number once, an agency quoted a round figure in a sales call — because nobody has actually observed Indian AI Overview citation patterns against real review data over time.

The decision this bears on is concrete: how a business paces its review-generation effort. A dental clinic with 40 reviews accumulating at 2 a month is a different investment case than one with 40 reviews accumulating at 15 a month, and if velocity turns out to matter independently of total count — which is RQ2 — that changes which of the two clinics should prioritise a review campaign right now versus wait. Getting this wrong in either direction wastes budget: over-investing in review volume once a business has already crossed whatever the real threshold is, or under-investing because a business assumes it needs to hit some large round number before AI citation becomes possible at all.

Study Design

Panel design: The study uses a longitudinal panel — the same businesses are observed monthly across 12 months. This design enables both cross-sectional comparison (threshold identification at any point) and longitudinal tracking (velocity and lag effect analysis).

Panel composition: 200 Indian businesses across:

• 6 categories: Healthcare (dermatology, IVF), Education (JEE coaching, general coaching), Restaurant, Hotel
• 4 cities: Bengaluru, Mumbai, Delhi NCR, Pune
• Review count distribution: intentionally stratified across the review count spectrum (0–50, 50–100, 100–150, 150–250, 250+) to capture threshold effects

Why a twelve-month panel, and why that's harder than a single audit

The reason nobody has published this kind of India-specific threshold data yet isn't that the idea is novel — it's that doing it properly requires exactly the kind of sustained, repeated observation that a one-time audit can't produce. A single snapshot showing which businesses with which review counts appear in AI Overviews right now tells you almost nothing about causation or stability. AI Overviews can cite a business one week and not the next for reasons unrelated to reviews entirely — a content update, a competing source getting crawled more recently, an algorithm tweak.

To separate a real threshold pattern from that kind of noise requires watching the same 200 businesses, monthly, for long enough to see seasonal cycles (festival seasons, exam seasons for the education category) play out at least once, and long enough to catch a meaningful number of businesses actually crossing from below-threshold to above-threshold during the study itself — which is the only way to measure the lag in RQ3 directly rather than infer it. That volume of repeated manual observation, done consistently by trained analysts rather than an unsupervised scraper, is the actual cost driver, and it's why this kind of data doesn't already exist for the Indian market.

Threshold Identification Methodology

For each category × city combination:

Step 1 — Monthly AI Overview observation: Run 10 standardised recommendation queries monthly ("best [category] in [city]"). Record which businesses appear in AI Overviews.

Step 2 — Review count recording: For each business that appears in AI Overviews, record current Google review count.

Step 3 — Threshold identification: The minimum review count observed among AI-cited businesses across the 12-month study period = the observed lower threshold bound.

The 25th percentile review count among AI-cited businesses = the "practical threshold" (25% of citations occur below this count; 75% above — reflecting that threshold varies and isn't a hard cutoff).

Why this methodology: Rather than attempting to reverse-engineer Google's algorithm (not possible), this approach observes outcomes empirically — what review counts are actually associated with AI citation. The resulting thresholds are descriptive (what is observed) rather than prescriptive (what Google requires).

Velocity Effect Measurement

Measuring independent velocity effect:

To isolate velocity's effect from total count, the study identifies "matched pairs" — business pairs where:

• Same category and city
• Total review count within 10% of each other
• Monthly velocity differs by at least 50% (e.g., one business receiving 8 new reviews/month, the matched business receiving 14+/month)

For matched pairs above the threshold, the study measures:

• AI Overview citation rate for the high-velocity business vs low-velocity business
• Citation consistency (% of months cited) for high-velocity vs low-velocity
• Citation description quality (positive/neutral/negative framing) for high-velocity vs low-velocity

Hypothesis: H1: High-velocity businesses above threshold receive AI Overview citations in a higher percentage of monthly observation periods than matched low-velocity businesses H2: High-velocity businesses receive more consistently positive citation framing

Lag Analysis: Threshold Crossing to Consistent Citation

For businesses crossing the review count threshold during the study period:

Businesses are identified that cross from below-threshold to above-threshold review counts during the 12-month study. The lag between threshold crossing and first AI Overview citation is measured.

Measurement:

• Date of threshold crossing: the month the business's review count first exceeds the category × city observed threshold
• Date of first AI Overview citation: the first monthly observation where the business appears in AI Overviews for any of the 10 standardised queries
• Lag = months between crossing and first citation

Secondary measurement: % of queries generating citation in months 1, 2, 3, and 6 after threshold crossing — measuring how citation consistency builds over time.

What would invalidate this study, and the confounders it can't fully remove

The matched-pair design for velocity exists specifically to answer a hard question: is it the number of reviews or the pace of reviews that matters, given that businesses accumulating reviews faster are also often businesses doing other things right at the same time — responding to customers well, running active marketing, maintaining a more complete GBP. Matching pairs on total review count controls for the most obvious confound, but it can't control for everything; a business with fast review velocity may also post to GBP more often, and this study's design doesn't fully separate posting frequency from review velocity as independent causes.

The bigger threat to validity is businesses that change materially during the study — renaming, relocating, or being acquired — which is why those cases get flagged and removed from the longitudinal analysis rather than left in to distort the threshold-crossing lag numbers. A second, more uncomfortable confound: review count jumps that look like organic velocity but might reflect review purchasing. The study can't verify authenticity directly, so sudden large jumps get flagged and analysed as a separate category rather than folded into the "high velocity" group, to avoid contaminating the legitimate velocity-effect estimate with purchased-review noise.

How a business can run a scaled-down version of this itself

You don't need a 200-business panel to learn something real about your own category's review dynamics. Pick your primary category + city query, run it in Google (checking for an AI Overview) once a week for six to eight weeks, and note every business that appears. For each one, record its current Google review count from Maps. Within a couple of months you'll have your own small, informal dataset showing roughly what review counts the businesses actually appearing in your specific local AI Overview currently hold — which tells you far more about your real competitive floor than any published national number would, because it's specific to your exact category and city.

If you want to check velocity rather than just count, do the same weekly check on your own listing and note your review count each week over the same period — if you cross into the range you observed as "typically cited" during the tracking window, watch whether you start appearing. That's not a controlled study, but it's a real, low-cost way to see whether the pattern in your category resembles what this research is designed to establish more rigorously at scale. Angryturtle's AI Search Readiness Audit and Review Velocity glossary page go into the mechanics of velocity measurement in more depth if you want to formalise the tracking.

What US-market review research doesn't tell us here

Review-count and review-velocity research tied to local search ranking has a long history in the US market, built around Google Reviews as effectively the only review signal that matters. That framing doesn't transfer directly to India, where a healthcare business's review footprint is split across Google and Practo, a restaurant's across Google and Zomato, and a general-services business's across Google and JustDial — and where AI systems may weight each source differently depending on which one they're actually able to browse for a given category. No specific US review-threshold figure is used anywhere in this methodology, because none has been verified as meaningful for an Indian category × city cell, and the multi-directory review landscape here is different enough that applying a US number would likely mislead rather than help.

Data Collection Protocols

GBP review count extraction: Monthly review counts are extracted from Google Maps search results for each business. Manual extraction is validated against GBP Insights data where available.

AI Overview observation: Standardised query protocol (same queries, same time of month, incognito browsing). Multiple observers for cross-validation on 15% of observations.

Data quality controls:

• Businesses that change name, close, or relocate during the study period are flagged and removed from longitudinal analysis
• Businesses that receive sudden large review count jumps (potential review purchasing) are flagged and analysed separately
• AI platform changes that may affect citation patterns independently of business-level signals are documented as study notes

Expected Outputs

The published study will report:

Threshold tables: Category × city minimum and practical thresholds based on observed data.

Velocity effect size: Standardised effect size (Cohen's d or equivalent) for velocity's independent contribution to AI citation rate above threshold.

Lag distribution: Median, mean, and interquartile range of the threshold-to-citation lag for each category.

Implications for investment sequencing: Based on observed thresholds and lags, recommendations for review building timeline expectations.

Limitations

AI system non-determinism: AI systems don't always produce the same response to the same query. Monthly measurement and multiple queries mitigate but don't eliminate this variance.

Confounds: Reviews are not the only variable — businesses crossing the threshold may also change GBP completeness, schema, or other signals during the study period. The matched-pair design for velocity analysis controls for some confounds but not all.

Evolving AI systems: AI Overview algorithms may change during the 12-month study. Such changes are documented in study notes but may affect comparability across periods.

How and when findings will publish

Category × city threshold tables, the velocity effect size, and the lag distribution will be published as part of Angryturtle's annual India AI Search Readiness Report once the full twelve-month panel completes and the matched-pair and lag analyses are validated. Aggregate tables are the public output; no individual business's review data or citation history is published. See the India AI Search Readiness Report for how this threshold work fits into the wider Reviews pillar of the readiness score.

FAQ Section

Q: Why 12 months? Can this research be done faster? A: The 12-month design captures seasonal variation in query patterns (festival seasons, exam seasons) and allows sufficient sample of threshold-crossing events for lag analysis. A 6-month study would reduce the threshold-crossing sample size significantly.

Q: Will this study be publicly available? A: Aggregate findings will be published as part of Angryturtle's India AI Search annual report. Category × city threshold tables are a key public output.

Q: Is there a single review count I should target right now? A: No — and any number offered to you as a universal threshold should be treated with scepticism. What matters is being competitive against the businesses currently ranking and getting cited in your own category and city, which is exactly what the scaled-down self-check above and the full AI Search Readiness Audit are both designed to show you.

See how this research applies to your business →

Internal links: AI Search Readiness Audit · Review Velocity glossary · Share of AI Voice Tracking · India AI Search Readiness Report · Reviews & AI Search Visibility blog · Review Sentiment glossary · Review Signals glossary · Learning Centre: Review Generation Engine · Learning Centre: Responding to Reviews · Learning Centre: Handling Fake & Negative Reviews · Product: Reviews AI · GBP Management Services · Vernacular Search & AI Citations study